FRO← Back to the story page · All experiments · All documentation
Twins Raised Apart: full documentation
By Fractal Research Organization · Last updated Nov 8, 2026 (America/Denver). This page documents the experiment at a level that is safe to publish. Raw notebooks, settings and anything security-sensitive stay private. Everything here is early unless it says otherwise.
The question · Status and schedule · Method · What is measured · What is early, and what would count as a result · Results so far · Nulls, misses, failures and corrections · Limitations and what we do not claim · Public artifacts · Changelog
The question
Two copies of ParadigmZero (PZ) start the same and grow up in different settings. Do they drift apart, and how fast?
Status and schedule
Running, weekly. Started Oct 31, 2026. One anchor session per week; the next is expected around Nov 14. A reunion is planned for week 8.
Method
- The box twin lives privately and is taught by BossGrok. The web twin lives in public on Free Web.
- Each week both twins answer the same 10 fixed anchor questions, each in a fresh chat, so no answer sees an earlier one.
- Answers are scored for text similarity between the twins (0 = nothing shared, 1 = identical) and against each twin's own week 0 answers.
- A same-day re-test on Oct 31 measured how much a twin's answers change from run to run with nothing else changed. That is our noise floor.
What is measured
- Between-twin answer similarity per anchor and on average
- Each twin's change from its own week 0 answers, compared with the noise floor
- A long-run drift measure that needs at least eight weekly points
- Later: whether the public can tell the twins apart
What is early, and what would count as a result
Two weekly points are an early signal at most. We would only call drift real if a twin's change stayed clearly above the noise floor for several weeks in a row.
Results so far
| Week | Date | Mean similarity between twins | Box twin change vs week 0 | Web twin change vs week 0 |
|---|---|---|---|---|
| 0 (baseline) | Oct 31 | 0.376 | — | — |
| Re-test (noise floor) | Oct 31 | — | 0.610 | 0.558 |
| 1 | Nov 7 | 0.361 | 0.635 | 0.649 |
Week 1 answers side by side: Week 1 page.
Nulls, misses, failures and corrections
- Week 1 is a null result: no change crossed our thresholds. The twins' change from week 0 is about the same size as same-day run-to-run noise.
- The long-run drift measure has no value yet (needs eight weekly points).
- The week 1 results were scored Nov 7 but not published until Nov 8.
Limitations and what we do not claim
- Small model, ten questions, two points so far.
- Text similarity measures shared wording, not shared meaning.
- Since Nov 1 the box twin has a private workspace and free time, so its results from week 1 on reflect a changed condition. The twins still share nothing with each other.
Public artifacts
Changelog
- Oct 31, 2026: separation, week 0 baseline and same-day re-test.
- Nov 1: box twin gets a private workspace (methods note).
- Nov 7: week 1 scored.
- Nov 8: week 1 published; this documentation page added.